快捷導(dǎo)航

Keras實現(xiàn)DenseNet結(jié)構(gòu)操作

更新時間：2020年07月06日 14:43:32 作者：小石學(xué)CS

這篇文章主要介紹了Keras實現(xiàn)DenseNet結(jié)構(gòu)操作，具有很好的參考價值，希望對大家有所幫助。一起跟隨小編過來看看吧

DenseNet結(jié)構(gòu)在16年由Huang Gao和Liu Zhuang等人提出，并且在CVRP2017中被評為最佳論文。網(wǎng)絡(luò)的核心結(jié)構(gòu)為如下所示的Dense塊，在每一個Dense塊中，存在多個Dense層，即下圖所示的H1～H4。各Dense層之間彼此均相互連接，即H1的輸入為x0，輸出為x1，H2的輸入即為[x0, x1]，輸出為x2，依次類推。最終Dense塊的輸出即為[x0, x1, x2, x3, x4]。這種結(jié)構(gòu)個人感覺非常類似生物學(xué)里邊的神經(jīng)元連接方式，應(yīng)該能夠比較有效的提高了網(wǎng)絡(luò)中特征信息的利用效率。

DenseNet的其他結(jié)構(gòu)就非常類似一般的卷積神經(jīng)網(wǎng)絡(luò)結(jié)構(gòu)了，可以參考論文中提供的網(wǎng)路結(jié)構(gòu)圖（下圖）。但是個人感覺，DenseNet的這種結(jié)構(gòu)應(yīng)該是存在進一步的優(yōu)化方法的，比如可能不一定需要在Dense塊中對每一個Dense層均直接進行相互連接，來縮小網(wǎng)絡(luò)的結(jié)構(gòu)；也可能可以在不相鄰的Dense塊之間通過簡單的下采樣操作進行連接，進一步提升網(wǎng)絡(luò)對不同尺度的特征的利用效率。

由于DenseNet的密集連接方式，在構(gòu)建一個相同容量的網(wǎng)絡(luò)時其所需的參數(shù)數(shù)量遠(yuǎn)小于其之前提出的如resnet等結(jié)構(gòu)。進一步，個人感覺應(yīng)該可以把Dense塊看做對一個有較多參數(shù)的卷積層的高效替代。因此，其也可以結(jié)合U-Net等網(wǎng)絡(luò)結(jié)構(gòu)，來進一步優(yōu)化網(wǎng)絡(luò)性能，比如單純的把U-net中的所有卷積層全部換成DenseNet的結(jié)構(gòu)，就可以顯著壓縮網(wǎng)絡(luò)大小。

下面基于Keras實現(xiàn)DenseNet-BC結(jié)構(gòu)。首先定義Dense層，根據(jù)論文描述構(gòu)建如下：

def DenseLayer(x, nb_filter, bn_size=4, alpha=0.0, drop_rate=0.2):
 
 # Bottleneck layers
 x = BatchNormalization(axis=3)(x)
 x = LeakyReLU(alpha=alpha)(x)
 x = Conv2D(bn_size*nb_filter, (1, 1), strides=(1,1), padding='same')(x)
 
 # Composite function
 x = BatchNormalization(axis=3)(x)
 x = LeakyReLU(alpha=alpha)(x)
 x = Conv2D(nb_filter, (3, 3), strides=(1,1), padding='same')(x)
 
 if drop_rate: x = Dropout(drop_rate)(x)
 
 return x

論文原文中提出使用1*1卷積核的卷積層作為bottleneck層來優(yōu)化計算效率。原文中使用的激活函數(shù)全部為relu，但個人習(xí)慣是用leakyrelu進行構(gòu)建，來方便調(diào)參。

之后是用Dense層搭建Dense塊，如下：

def DenseBlock(x, nb_layers, growth_rate, drop_rate=0.2):
 
 for ii in range(nb_layers):
  conv = DenseLayer(x, nb_filter=growth_rate, drop_rate=drop_rate)
  x = concatenate([x, conv], axis=3)
 return x

如論文中所述，將每一個Dense層的輸出與其輸入融合之后作為下一Dense層的輸入，來實現(xiàn)密集連接。

最后是各Dense塊之間的過渡層，如下：

def TransitionLayer(x, compression=0.5, alpha=0.0, is_max=0):
 
 nb_filter = int(x.shape.as_list()[-1]*compression)
 x = BatchNormalization(axis=3)(x)
 x = LeakyReLU(alpha=alpha)(x)
 x = Conv2D(nb_filter, (1, 1), strides=(1,1), padding='same')(x)
 if is_max != 0: x = MaxPooling2D(pool_size=(2, 2), strides=2)(x)
 else: x = AveragePooling2D(pool_size=(2, 2), strides=2)(x)
 
 return x

論文中提出使用均值池化層來作下采樣，不過在邊緣特征提取方面，最大池化層效果應(yīng)該更好，這里就加了相關(guān)接口。

將上述結(jié)構(gòu)按照論文中提出的結(jié)構(gòu)進行拼接，這里選擇的參數(shù)是論文中提到的L=100，k=12，網(wǎng)絡(luò)連接如下：

growth_rate = 12
inpt = Input(shape=(32,32,3))
 
x = Conv2D(growth_rate*2, (3, 3), strides=1, padding='same')(inpt)
x = BatchNormalization(axis=3)(x)
x = LeakyReLU(alpha=0.1)(x)
x = DenseBlock(x, 12, growth_rate, drop_rate=0.2)
x = TransitionLayer(x)
x = DenseBlock(x, 12, growth_rate, drop_rate=0.2)
x = TransitionLayer(x)
x = DenseBlock(x, 12, growth_rate, drop_rate=0.2)
x = BatchNormalization(axis=3)(x)
x = GlobalAveragePooling2D()(x)
x = Dense(10, activation='softmax')(x)
 
model = Model(inpt, x)
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.summary()

雖然我們已經(jīng)完成了網(wǎng)絡(luò)的架設(shè)，網(wǎng)絡(luò)本身的參數(shù)數(shù)量也僅有0.5M，但由于以這種方式實現(xiàn)的網(wǎng)絡(luò)在Dense塊中，每一次concat均需要開辟一組全新的內(nèi)存空間，導(dǎo)致實際需要的內(nèi)存空間非常大。作者在17年的時候，還專門寫了相關(guān)的技術(shù)報告：https://arxiv.org/abs/1707.06990來說明怎么節(jié)省內(nèi)存空間，不過單純用keras實現(xiàn)起來是比較麻煩。下一篇博客中將以pytorch框架來對其進行實現(xiàn)。

最后放出網(wǎng)絡(luò)完整代碼：

import numpy as np
import keras
from keras.models import Model, save_model, load_model
from keras.layers import Input, Dense, Dropout, BatchNormalization, LeakyReLU, concatenate
from keras.layers import Conv2D, MaxPooling2D, AveragePooling2D, GlobalAveragePooling2D
 
## data
import pickle
 
data_batch_1 = pickle.load(open("cifar-10-batches-py/data_batch_1", 'rb'), encoding='bytes')
data_batch_2 = pickle.load(open("cifar-10-batches-py/data_batch_2", 'rb'), encoding='bytes')
data_batch_3 = pickle.load(open("cifar-10-batches-py/data_batch_3", 'rb'), encoding='bytes')
data_batch_4 = pickle.load(open("cifar-10-batches-py/data_batch_4", 'rb'), encoding='bytes')
data_batch_5 = pickle.load(open("cifar-10-batches-py/data_batch_5", 'rb'), encoding='bytes')
 
train_X_1 = data_batch_1[b'data']
train_X_1 = train_X_1.reshape(10000, 3, 32, 32).transpose(0, 2, 3, 1).astype("float")
train_Y_1 = data_batch_1[b'labels']
 
train_X_2 = data_batch_2[b'data']
train_X_2 = train_X_2.reshape(10000, 3, 32, 32).transpose(0, 2, 3, 1).astype("float")
train_Y_2 = data_batch_2[b'labels']
 
train_X_3 = data_batch_3[b'data']
train_X_3 = train_X_3.reshape(10000, 3, 32, 32).transpose(0, 2, 3, 1).astype("float")
train_Y_3 = data_batch_3[b'labels']
 
train_X_4 = data_batch_4[b'data']
train_X_4 = train_X_4.reshape(10000, 3, 32, 32).transpose(0, 2, 3, 1).astype("float")
train_Y_4 = data_batch_4[b'labels']
 
train_X_5 = data_batch_5[b'data']
train_X_5 = train_X_5.reshape(10000, 3, 32, 32).transpose(0, 2, 3, 1).astype("float")
train_Y_5 = data_batch_5[b'labels']
 
train_X = np.row_stack((train_X_1, train_X_2))
train_X = np.row_stack((train_X, train_X_3))
train_X = np.row_stack((train_X, train_X_4))
train_X = np.row_stack((train_X, train_X_5))
 
train_Y = np.row_stack((train_Y_1, train_Y_2))
train_Y = np.row_stack((train_Y, train_Y_3))
train_Y = np.row_stack((train_Y, train_Y_4))
train_Y = np.row_stack((train_Y, train_Y_5))
train_Y = train_Y.reshape(50000, 1).transpose(0, 1).astype("int32")
train_Y = keras.utils.to_categorical(train_Y)
 
test_batch = pickle.load(open("cifar-10-batches-py/test_batch", 'rb'), encoding='bytes')
test_X = test_batch[b'data']
test_X = test_X.reshape(10000, 3, 32, 32).transpose(0, 2, 3, 1).astype("float")
test_Y = test_batch[b'labels']
test_Y = keras.utils.to_categorical(test_Y)
 
train_X /= 255
test_X /= 255
 
# model
 
def DenseLayer(x, nb_filter, bn_size=4, alpha=0.0, drop_rate=0.2):
 
 # Bottleneck layers
 x = BatchNormalization(axis=3)(x)
 x = LeakyReLU(alpha=alpha)(x)
 x = Conv2D(bn_size*nb_filter, (1, 1), strides=(1,1), padding='same')(x)
 
 # Composite function
 x = BatchNormalization(axis=3)(x)
 x = LeakyReLU(alpha=alpha)(x)
 x = Conv2D(nb_filter, (3, 3), strides=(1,1), padding='same')(x)
 
 if drop_rate: x = Dropout(drop_rate)(x)
 
 return x
 
def DenseBlock(x, nb_layers, growth_rate, drop_rate=0.2):
 
 for ii in range(nb_layers):
  conv = DenseLayer(x, nb_filter=growth_rate, drop_rate=drop_rate)
  x = concatenate([x, conv], axis=3)
  
 return x
 
def TransitionLayer(x, compression=0.5, alpha=0.0, is_max=0):
 
 nb_filter = int(x.shape.as_list()[-1]*compression)
 x = BatchNormalization(axis=3)(x)
 x = LeakyReLU(alpha=alpha)(x)
 x = Conv2D(nb_filter, (1, 1), strides=(1,1), padding='same')(x)
 if is_max != 0: x = MaxPooling2D(pool_size=(2, 2), strides=2)(x)
 else: x = AveragePooling2D(pool_size=(2, 2), strides=2)(x)
 
 return x
 
growth_rate = 12
 
inpt = Input(shape=(32,32,3))
 
x = Conv2D(growth_rate*2, (3, 3), strides=1, padding='same')(inpt)
x = BatchNormalization(axis=3)(x)
x = LeakyReLU(alpha=0.1)(x)
x = DenseBlock(x, 12, growth_rate, drop_rate=0.2)
x = TransitionLayer(x)
x = DenseBlock(x, 12, growth_rate, drop_rate=0.2)
x = TransitionLayer(x)
x = DenseBlock(x, 12, growth_rate, drop_rate=0.2)
x = BatchNormalization(axis=3)(x)
x = GlobalAveragePooling2D()(x)
x = Dense(10, activation='softmax')(x)
 
model = Model(inpt, x)
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
 
model.summary()
 
for ii in range(10):
 print("Epoch:", ii+1)
 model.fit(train_X, train_Y, batch_size=100, epochs=1, verbose=1)
 score = model.evaluate(test_X, test_Y, verbose=1)
 print('Test loss =', score[0])
 print('Test accuracy =', score[1])
 
save_model(model, 'DenseNet.h5')
model = load_model('DenseNet.h5')
 
pred_Y = model.predict(test_X)
score = model.evaluate(test_X, test_Y, verbose=0)
print('Test loss =', score[0])
print('Test accuracy =', score[1])

以上這篇Keras實現(xiàn)DenseNet結(jié)構(gòu)操作就是小編分享給大家的全部內(nèi)容了，希望能給大家一個參考，也希望大家多多支持腳本之家。

您可能感興趣的文章: